Google Trends Scraper

Get Google Trends top and rising related queries, interest over time and interest by country, region, metro area or city for any keyword, or compare up to 5 keywords. Pick any location, time range (past hour to 2004–today, or custom dates), category and search type (web, images, news, shopping, YouTube).

You pay only for keywords whose interest over time has data. Keywords with too little search volume, inputs that can't be sent, and searches Google rate-limits are free rows with the reason. A charged row can still have an empty region list or empty related-query lists: when Google has none for that keyword (small keywords, city level), or when Google kept rate-limiting or refused that one part (then sectionErrors says so).

What you get

One row per keyword per search:

Field	What it is
keyword, comparedWith	The keyword and the other keywords in its comparison
interestOverTime	date (UTC), value (0–100, relative to the peak in the search), formattedValue (as the website shows it: "<1" when the value is 0 but there is data), isPartial (the last, still-running period)
averageInterest	Average over the period (comparisons only; Google gives none for single keywords)
interestByRegion	geoCode, geoName, value (0–100) and formattedValue ("<1" where the value is 0) for each place with data, highest first. City level gives coordinates instead of a code
relatedQueries	top and rising lists: query, value and formattedValue (rising: % growth, or "Breakout")
sectionErrors	Only when a region list or related queries could not be fetched although the interest over time could: which part, and why
status	ok (charged), or no_data, invalid, rejected, blocked, error (free, with error saying why)

Input

• Search terms: one search per line. coffee is one search; coffee, tea, water compares three keywords on one scale (values are relative to the highest of them). A comma always means compare: Washington, DC is two keywords and two results; write Washington DC. Empty lines are skipped.

• Location: a country or region code (US, GB, US-CA); empty for worldwide. UK is read as GB.

• Region level must be finer than the location (a country: regions, metro areas or cities; a US state: metro areas or cities).

• Time range: past hour, 4 hours, day, 7 days, 30 days, 90 days, 12 months, 5 years, since 2004, custom dates (2024-01-01 2024-06-30), or custom hours in UTC (2026-10-09T10 2026-10-10T22). Upper or lower case both work.

• Category ID and search type narrow the search, as on the Trends website.

• Sections: switch interest by region or related queries off if you don't need them; runs get faster.

• Max concurrency: searches fetched in parallel, 5 by default, at most 8. Above that Google rate-limits harder and more searches end up blocked.

Good to know

• Google Trends values are relative (0–100), not search counts, and relative to each search: the same keyword gives different numbers in a different comparison, region or time range.
• Values come from the Trends API that the website itself uses, as Google answers automated requests.
• Google Trends is a sample with added noise. Each request can get a different sample (it changes by day and by the requesting location), so values can differ by a few points from the website, and more for small keywords or regions. For research, average runs from different days.
• Values from different lines are on different scales. To compare more than 5 keywords, repeat one well-searched anchor keyword in every line and rescale by it.
• Ranges up to 7 days are in UTC here; the website shows your local time. Weekly points are labelled by the Sunday that starts the week. value is empty (null) for periods without data.
• Region values are each place's share of its own searches, so a small place can top a worldwide list; formattedValue "<1" marks a 0 that means "below 1". Long ranges show Google's own method changes, e.g. a jump on 1 January 2016.
• Rising related queries sometimes contain generic phrases unrelated to the keyword (e.g. "top songs this week" for coffee). The Trends website shows the same ones; they are in Google's data.
• In a comparison, a keyword far smaller than the others shows "<1" (value 0) for most periods, exactly as on the website. It has data, so it is a charged result.
• Filter your results on status = ok to keep only rows with data.
• Spreadsheets: the default export puts every date in its own column. The dataset views Interest over time and Interest by region give one row per date or per place instead. Through the API, add fields=keyword,geo,interestOverTime&unwind=interestOverTime (or interestByRegion) to the dataset items URL.
• Run outcome: the OUTPUT record in the run's key-value store says whether the run succeeded, with its final message, the number of keyword results, counts per status and searches not processed. Handy for Make, Zapier, n8n and AI agents.
• Long lists: a list over ~50 keywords takes longer than 5 minutes, so start it as a normal (asynchronous) run; Apify's synchronous API endpoint gives up after 300 seconds. The default run timeout is 24 hours, and the actor stops cleanly before any timeout, telling you how many searches were not processed. Google's rate limits set the pace: one run gets through roughly 10,000–20,000 searches in 24 hours, so split bigger lists into several runs. One run takes at most 50,000 searches.
• Google rate-limits Trends heavily. The actor paces its requests, rotates IPs and retries, and gives searches that were still rate-limited one more try at the end of the run; a search that still fails is a free blocked row, and a run where most searches fail ends as failed so you can retry later. On Apify's free plan your runs share fewer proxy IPs and see more rate-limited (free) rows.
• Metro areas (DMA) exist only for the United States.
• Related topics are not included.
• This actor reads public, aggregated Trends data only. No personal data.

Pricing

Pay per result: one charge per keyword whose interest over time has data, plus a small charge per run start. Rows without data are free. A 5-keyword comparison is 5 results; there is no platform usage on top. Set a maximum charge per run in the run options; the actor stops cleanly when it is reached.